app.py

import gradio as gr
import fitz  # PyMuPDF
import os
from sentence_transformers import SentenceTransformer
import numpy as np
import faiss
import uuid
from groq import Groq

# Load embedding model
embedder = SentenceTransformer("all-MiniLM-L6-v2")

# Initialize vector store and document store
document_chunks = []
doc_embeddings = []
doc_ids = []
index = None

# Get Groq API key from environment variable
GROQ_API_KEY = os.getenv("GROQ_API_KEY")

client = Groq(api_key=GROQ_API_KEY)

# Load and split PDF
def extract_text_from_pdf(pdf_path):
    doc = fitz.open(pdf_path)
    text = ""
    for page in doc:
        text += page.get_text()
    return text

# Chunking logic
def chunk_text(text, max_tokens=500):
    import re
    sentences = re.split(r'(?<=[.!?]) +', text)
    chunks = []
    chunk = ""
    tokens = 0

    for sentence in sentences:
        sentence_tokens = len(sentence.split())
        if tokens + sentence_tokens > max_tokens:
            chunks.append(chunk.strip())
            chunk = sentence
            tokens = sentence_tokens
        else:
            chunk += " " + sentence
            tokens += sentence_tokens

    if chunk:
        chunks.append(chunk.strip())
    return chunks

# Indexing
def index_pdf(pdf_file):
    global document_chunks, doc_embeddings, doc_ids, index

    if not pdf_file:
        return "❌ Please upload a PDF file."

    text = extract_text_from_pdf(pdf_file.name)
    document_chunks = chunk_text(text)

    doc_embeddings = embedder.encode(document_chunks)
    doc_embeddings = np.array(doc_embeddings).astype("float32")

    dimension = doc_embeddings.shape[1]
    index = faiss.IndexFlatL2(dimension)
    index.add(doc_embeddings)

    doc_ids = [str(uuid.uuid4()) for _ in range(len(document_chunks))]

    return "✅ PDF indexed successfully. You can now ask questions."

# Retrieve top chunks
def retrieve_relevant_chunks(query, k=3):
    query_embedding = embedder.encode([query]).astype("float32")
    distances, indices = index.search(query_embedding, k)
    return [document_chunks[i] for i in indices[0]]

# Generate answer using Groq
def generate_answer(user_query):
    if index is None:
        return "❌ Please upload and index a PDF first."

    top_chunks = retrieve_relevant_chunks(user_query, k=3)
    context = "\n\n".join(top_chunks)

    messages = [
        {"role": "system", "content": "You are a helpful academic assistant who answers questions based on uploaded PDF papers."},
        {"role": "user", "content": f"Context: {context}\n\nQuestion: {user_query}"}
    ]

    try:
        response = client.chat.completions.create(
            messages=messages,
            model="llama3-8b-8192",
        )
        return response.choices[0].message.content.strip()
    except Exception as e:
        return f"❌ Error generating response: {e}"

# Gradio UI
with gr.Blocks(title="📘 PDF Question Assistant") as demo:
    gr.Markdown("# 📘 Ask Questions About Your PDF")
    with gr.Tab("📄 Upload & Index"):
        with gr.Row():
            pdf_input = gr.File(label="Upload PDF File", type="filepath", file_types=[".pdf"])
            upload_btn = gr.Button("🔍 Index PDF", variant="primary")
        upload_status = gr.Textbox(label="", interactive=False, placeholder="Status will appear here...")

    with gr.Tab("❓ Ask a Question"):
        with gr.Row():
            query = gr.Textbox(label="Ask something from the PDF", placeholder="E.g. What is the main argument of the paper?")
            query_btn = gr.Button("🧠 Get Answer")
        answer = gr.Textbox(label="Answer", placeholder="AI-generated answer will appear here...", lines=8)

    upload_btn.click(fn=index_pdf, inputs=[pdf_input], outputs=[upload_status])
    query_btn.click(fn=generate_answer, inputs=[query], outputs=[answer])

if __name__ == "__main__":
    demo.launch()